The most useful colonoscopy AI result in a while, because the target is the operator rather than the image. Endoscopists received computer-aided quality feedback after each procedure and got measurably better: adenoma detection rose from 43.4% to 48.6%, and — the number that matters — adenocarcinoma detection rose from 6.5% to 8.9%. Two features make this more persuasive than the usual CADe trial. The 17 endoscopists acted as their own controls in a stepped-wedge design, and the gain did not depend on a device being switched on during the case, so it is a change in practice rather than a change in equipment. Read the effect carefully, though: adenomas per colonoscopy did not reach significance, so this is more patients with a lesion found, not more lesions per patient. If you run a screening programme, this is a cheap intervention with a hard endpoint attached.
· ADR 48.6% vs 43.4% (OR 1.24, 95% CI 1.02–1.52, p=0.032)· Adenocarcinoma detection 8.9% vs 6.5% (OR 1.51, 95% CI 1.06–2.15, p=0.027)· Polyp detection 54.2% vs 50.7% (OR 1.22, 95% CI 1.00–1.49, p=0.048)· Withdrawal time 960 vs 915 s (IRR 1.11, 95% CI 1.06–1.17, p=0.003)· Adenomas per colonoscopy 1.00 vs 0.85 (IRR 1.15, 95% CI 0.99–1.33, p=0.069)· 2110 FIT-positive patients aged 50–74, 17 endoscopists, 3 Danish university hospitals
Adenoma detection rate · 0–100% scale
48.6%With CAQ feedback
43.4%Before feedback
050100%
Go deeper
A stepped-wedge cluster-randomised trial across three Danish university hospitals tested computer-aided quality (CAQ) feedback delivered to endoscopists after the procedure — not computer-aided detection during it. Among 2110 FIT-positive patients aged 50–74, with 17 endoscopists acting as their own controls across three clusters, adenoma detection rate rose from 43.4% to 48.6% (OR 1.24, 95% CI 1.02–1.52, p=0.032), and adenocarcinoma detection rate from 6.5% to 8.9% (OR 1.51, 95% CI 1.06–2.15, p=0.027). Polyp detection rate improved marginally (54.2% vs 50.7%, OR 1.22, 95% CI 1.00–1.49, p=0.048) and withdrawal time lengthened from 915 to 960 seconds (IRR 1.11, 95% CI 1.06–1.17, p=0.003). Adenomas per colonoscopy did not reach significance (1.00 vs 0.85, IRR 1.15, 95% CI 0.99–1.33, p=0.069), so the effect is better read as more patients with a lesion found than more lesions per patient. This is a different intervention from the CADe trials this digest has covered — the feedback loop targets the operator, and the gain persisted rather than depending on a device being switched on. Full daily entry →
Germline multigene panel testing in colorectal cancer
The argument for age-restricted germline testing in colorectal cancer gets harder to sustain after this. Pooled across 21 studies and 6,925 patients, 15.2% carried a pathogenic or likely pathogenic variant. The finding that should change behaviour is not the early-onset number, high as it is, but the late-onset one: above age 50 the any-variant yield was still 13.0% and high-penetrance 6.5%. Modelling put the age at which high-penetrance yield finally drops below the conventional 5% actionability threshold at 67 — decades beyond where most criteria stop. Neither panel size nor publication year explained the yield, which argues this is not an artefact of newer, larger panels. Heterogeneity was substantial (I²=84.1%) and two studies were at high risk of bias, so treat the point estimate as approximate. The direction, though, is not ambiguous.
· Any P/LP variant 15.2% (95% CI 12.8–18.1; I²=84.1%)· High-penetrance variants 7.9% (95% CI 5.9–10.5)· High-penetrance CRC-predisposition genes 5.4% (95% CI 3.5–8.6)· Early-onset <50y: any 17.7%, high-penetrance 14.2%· Late-onset ≥50y: any 13.0% (95% CI 10.8–15.7), high-penetrance 6.5% (95% CI 4.4–9.5)· High-penetrance yield stays above 5% to age 67 (95% CI 59–75)· 21 studies, 6,925 individuals· Risk of bias low in 10, moderate in 9, high in 2
Any pathogenic / likely pathogenic variant, by age at onset · 0–100% scale
17.7%Early-onset (<50y)
13.0%Late-onset (≥50y)
050100%
Go deeper
A PROSPERO-registered systematic review and meta-analysis of 21 studies and 6,925 individuals with colorectal cancer puts the pooled yield of germline multigene panel testing at 15.2% (95% CI 12.8–18.1; I²=84.1%) for any pathogenic or likely pathogenic variant, with high-penetrance variants in 7.9% (5.9–10.5) and high-penetrance CRC-predisposition genes in 5.4% (3.5–8.6). Yield was highest in early-onset disease — 17.7% any and 14.2% high-penetrance below age 50 — but the number that matters for practice is what happens above that threshold: in late-onset CRC the any-variant yield was still 13.0% (10.8–15.7) and high-penetrance 6.5% (4.4–9.5). Modelling put the age at which high-penetrance yield finally falls below 5% at 67 years (95% CI 59–75), well beyond the cutoffs most testing criteria still use. Neither panel size nor publication year affected yield, which argues the result is not an artifact of newer and larger panels. Risk of bias was low in 10 studies, moderate in 9 and high in 2, and heterogeneity was substantial. Full daily entry →
A decade of real-world data, and the lesson is not about the test. Mayo's programme ran 148,624 multi-target stool DNA tests, and the diagnostic yield after a positive result was genuinely strong — advanced adenoma in 30.7%, adenocarcinoma in 1.3%, with an adenoma detection rate above society benchmarks. The failure is downstream: only 65.1% of 19,506 positive results led to a completed colonoscopy within a year, and longer delays tracked with higher advanced adenoma rates. Adherence after a negative test was worse — 35.3% re-screened within three years — and colonoscopy done within three years of a negative test still found an adenoma or serrated lesion in 45.2%. A positive stool DNA test is not a screening result, it is an appointment that has not happened yet. Worth auditing your own completion rate before adding any new modality. One methodological caveat to name rather than bury: endpoints were extracted from free-text reports by a validated large-language-model pipeline.
· 148,624 tests in 110,410 individuals (Sept 2014 – July 2025)· Positivity 13.6%· Colonoscopy completed within 12 months of a positive result: 65.1% of 19,506 (median 82 days)· Yield after a positive: advanced adenoma 30.7%, adenocarcinoma 1.3%· ADR after a positive >50%, exceeding society benchmarks and rising over the decade· Re-screened within 3 years of a negative: 35.3%· Colonoscopy within 3 years of a negative still found adenoma or serrated lesion 45.2%, advanced adenoma 9.3%, cancer 0.5%· Longer interval to colonoscopy associated with higher advanced adenoma rate (trend p=0.001)
Follow-through after a multi-target stool DNA test · 0–100% scale
65.1%Colonoscopy ≤12 mo after a positive
35.3%Re-screened ≤3 yr after a negative
050100%
Go deeper
Mayo's decade-long program covers 148,624 multi-target stool DNA tests in 110,410 individuals between September 2014 and July 2025, with a 13.6% positivity rate. The diagnostic yield after a positive test was strong — advanced adenoma in 30.7% and adenocarcinoma in 1.3%, with an adenoma detection rate above 50% that exceeded society benchmarks and rose over the decade. The failure sits downstream of the assay: of 19,506 positive results, only 65.1% led to a completed colonoscopy within 12 months (median 82 days), and longer intervals to colonoscopy were associated with higher advanced adenoma rates (trend p=0.001). Adherence after a negative test was worse still, with only 35.3% re-screened within three years — and colonoscopy performed within three years of a negative mt-sDNA still found adenomas or serrated lesions in 45.2%, advanced adenoma in 9.3% and cancer in 0.5%. Endpoints were extracted from free-text reports by a validated large-language-model pipeline, which is worth naming as a methodological caveat rather than burying in the methods. Full daily entry →
A well-designed negative trial, and worth reading precisely because the negative means something. ANCHOR randomised 748 patients with RAS/BRAF wild-type metastatic colorectal cancer to first-line anlotinib or bevacizumab, each with chemotherapy. Median PFS was identical at 11.0 months, hazard ratio 1.00 — but the confidence interval crosses the prespecified noninferiority margin of 1.09, so noninferiority was not demonstrated and the trial does not license substituting the oral TKI for the established backbone. Meanwhile grade ≥3 treatment-related adverse events ran 64.9% versus 44.8%. The paper describes 'similar antitumor activity', which is a fair description of the point estimate and not a demonstration of equivalence — a distinction worth holding onto, because identical medians are exactly the situation where a failed noninferiority test is easiest to misreport as a positive one.
· Median PFS 11.0 vs 11.0 months, stratified HR 1.00 (95% CI 0.84–1.18), p=0.87· Prespecified noninferiority margin HR 1.09 — the CI crosses it, so noninferiority was NOT met· Grade ≥3 treatment-related adverse events 64.9% vs 44.8%· 748 patients randomised 1:1, anlotinib (n=373) vs bevacizumab (n=375), each with chemotherapy· Median follow-up 25.1 months (95% CI 23.8–26.3)· NCT04854668 / CTR20210940
HR 1No significant difference95% CI 0.84–1.18 · Δ +0 mo median · p=0.87
ANCHOR randomised 748 patients with RAS/BRAF wild-type metastatic colorectal cancer 1:1 to first-line anlotinib (n=373) or bevacizumab (n=375), each with chemotherapy, at multiple Chinese centres (NCT04854668 / CTR20210940). At a median follow-up of 25.1 months (95% CI 23.8–26.3), median PFS was identical at 11.0 months in both arms, with a stratified HR of 1.00 (95% CI 0.84–1.18, p=0.87). The prespecified noninferiority margin was HR 1.09, and the confidence interval crosses it — so noninferiority was not met, and the trial does not license substituting anlotinib for bevacizumab. Grade ≥3 treatment-related adverse events were substantially more common with anlotinib (64.9% vs 44.8%). The paper's phrase 'similar antitumor activity' describes the point estimate, not a demonstrated equivalence, and the honest read is that an oral TKI failed to replace an established backbone while adding 20 points of high-grade toxicity. Full daily entry →
An ancillary analysis of a randomised phase 3 that unsettles a threshold everyone uses. Scoring PD-L1 in 540 patients who received chemotherapy without any checkpoint inhibitor, CPS ≥5 showed no association with overall or progression-free survival, while TPS ≥1% was associated with both. Keep the interpretation tight: because nobody in this cohort received immunotherapy, the finding is prognostic, not predictive. It says something about tumour biology, not about who should be offered a checkpoint inhibitor, and it is not a reason to change how you select patients for one. What it does suggest is that CPS and TPS are not interchangeable readouts of the same thing, and that the prognostic weight sits with tumour-cell staining rather than the combined score. Worth watching as the assay-threshold literature in gastric cancer continues to fragment.
· CPS ≥5 — OS HR 0.86 (95% CI 0.71–1.04, p=0.127), PFS HR 0.92 (95% CI 0.77–1.11, p=0.392): not associated· TPS ≥1% — OS HR 0.78 (95% CI 0.62–0.97, p=0.028), PFS HR 0.76 (95% CI 0.62–0.94, p=0.012)· Dako 28-8 pharmDx assay in 540 of 741 enrolled patients· All patients received chemotherapy without an immune checkpoint inhibitor· Ancillary analysis of the randomised phase 3 JCOG1013
Go deeper
This ancillary analysis of the randomised Phase III JCOG1013 trial scored PD-L1 with the Dako 28-8 pharmDx assay in 540 of the 741 enrolled patients, all of whom received chemotherapy without an immune checkpoint inhibitor. The two scoring systems diverged. CPS ≥5 — the threshold that drives first-line ICI eligibility in gastric cancer — was not associated with overall survival (HR 0.86, 95% CI 0.71–1.04, p=0.127) or progression-free survival (HR 0.92, 0.77–1.11, p=0.392). TPS ≥1% was associated with both: OS HR 0.78 (0.62–0.97, p=0.028) and PFS HR 0.76 (0.62–0.94, p=0.012). The authors conclude that TPS ≥1% is an independent favourable prognostic factor on chemotherapy alone. The essential caveat is the population: because nobody here received immunotherapy, the finding is prognostic and not predictive of ICI benefit. It says something about tumour biology, not about who should be offered a checkpoint inhibitor. Full daily entry →
* Survival curves are illustrative — modeled from each trial’s reported median and hazard ratio (assuming exponential survival), not the published Kaplan–Meier curves.
The take
The take
Not one new drug earned a place in this issue, and the week was still a good one. The only large randomised therapeutic trial — ANCHOR, 748 patients — returned a hazard ratio of exactly 1.00 and failed its noninferiority test. Everything else that mattered was about selection and follow-through: who gets tested, at what age, on which threshold, and whether anything actually happens after the result comes back.
Read together, four of these five findings say the same uncomfortable thing — the decision rules in routine use are drawn in the wrong place:
Germline testing criteria stop decades before the yield does. High-penetrance yield holds above 5% to age 67.
CPS ≥5 carried no prognostic signal on chemotherapy; TPS ≥1% did. The two scores are not interchangeable.
A positive stool DNA test reached a colonoscopy within a year only 65% of the time. The assay was never the weak link.
The endoscopist, given feedback, detected more cancers. No new device required.
Two graded documents landed in the same week and belong on the same shelf as the papers above: the ACG clinical guideline on adenomatous colorectal polyposis syndromes, which covers germline testing selection and the extracolonic sites that get forgotten, and the ESGE 2026 guideline on biliary strictures, whose malignant-obstruction recommendations govern how a pancreatic or biliary cancer patient gets decompressed before systemic therapy starts. Both are in the daily editions linked from each card.
The unglamorous half of this specialty — who we test, how well we scope, whether the follow-up happens — moved further this week than the therapeutics did. That is worth sitting with.
GI Oncology Weekly · curated by Dr. Allan Pereira, Moffitt Cancer Center. Educational summaries for practising oncologists — not medical advice; verify against primary sources before acting.